HLS Foundation model in Google Cloud Platform Vertex AI

This repo deploys the recently published finetuned models based on Harmonized Landsat and Sentinel-2 (HLS) into Google Cloud Platform (GCP)'s Vertex AI. We host the models in Vertex AI's endpoint.

Steps to deploy:

Note: These steps are also presented in the notebook

1. Clone this repository git clone https://github.com/nasa-impact/hls-foundation-vertexai.git
2. Change directory into the cloned repository cd hls-foundation-vertexai
3. Initialize gcloud
4. Install required packages: ``` # Required in Docker serving container ! pip3 install -U -r requirements.txt -q --user

For local FastAPI development and running

! pip3 install -U "uvicorn[standard]>=0.12.0,<0.14.0" fastapi~=0.63 -q --user

Vertex SDK for Python

! pip3 install --upgrade --quiet google-cloud-aiplatform --user 5. Check to see if gcloud configuration is done properly.

List configuration

! gcloud config list

List projects

! gcloud projects list 6. Use GCP code build to create and push new artifact to be used in vertex AI

Replace with one of the project ids from above.

! gcloud builds submit --region=us-central1 --tag=us-central1-docker.pkg.dev//hls-foundation-vertexai/inference 7. Register artifact as model in vertex AI (Please use notebooks/colab from here on) from google.cloud import aiplatform

model = aiplatform.Model.upload( displayname='hls-inference', servingcontainerimageuri="us-central1-docker.pkg.dev//hls-foundation-vertexai/inference", ) 8. Create new vertex AI endpoint endpoint = model.deploy(machinetype="n1-standard-4", acceleratortype='NVIDIATESLAV100', acceleratorcount=1) endpoint.todict()['deployedModels'][0]['id'] 9. Create test data %%writefile test.json { "instances":{"date":"2023-08-13","boundingbox":[-156.81605703476012,20.69675592885614,-156.41605703476014,21.096755928856137], "modelid": "burn_scars"}, }

``` 10. Test endpoint

Note: Replace all with your project id, and with the output from step 8.

```

Get inference from the deployed endpoint. Copy over the endpoint id from above and replace , and project id from about to replace

! export ENDPOINTID=<ENDPOINTID>; export PROJECTID=<PROJECTID>; export INPUTDATAFILE="test.json"; curl \ -X POST \ -H "Authorization: Bearer $(gcloud auth print-access-token)" \ -H "Content-Type: application/json" \ https://us-central1-aiplatform.googleapis.com/v1/projects/${PROJECTID}/locations/us-central1/endpoints/${ENDPOINTID}:predict \ -d "@${INPUTDATAFILE}" ```